Comment by DeathArrow

8 hours ago

Do all llm know they are a LLM? It doesn't depend on the system prompt?

The pre-trained ones no (except some of the new ones which have post training data added to pre-training for some reason). The post-trained ones yes (at least all the ones I've seen).

Some of the niche ones I'm not sure about. Like the historical LLMs. I have not tested those yet.

Yes. The first step of aligning each and every GPT-based LLM is to suppress the “I am human” kind of responses. It’s baked into the weights.

  • Reminds me of old cleverbot conversations where it would always assert it is human and you are the bot.

    Trained on previous conversations with people.

  • It's also at minimum baked into the system prompt of virtually any LLM.

    • That's not "baked" and only applies to remotely hosted LLMs where someone else feeds the prompt into the LLM.

Without a system prompt no. And in general they “know” nothing and just predict the next best word.

  • For sure, as they are stochastic parrots. My question should have been: what are the odds a llm would react properly to those instruction, but I got lazy and asked if they "know" it, because I presumed most readers here do know how llms are working.